25 Fields Medal laureates published a joint statement in which they directly point to a "serious divergence of goals" between companies developing AI and the mathematical community. The document appeared on 11 September 2026 in Le Monde and on Terence Tao's blog; among the signatories are Pierre Deligne, Peter Scholze, Maryna Viazovska, Manjul Bhargava, Yu Deng and others. The statement arose against the backdrop of recent OpenAI announcements about solving the Navier–Stokes problem in 88 hours using thousands of agents and millions of messages.
The essence of the complaint is not that AI solves problems, but in precisely how this happens and what values are thereby promoted. Mathematics has traditionally valued not the mere fact of a solution, but the understanding of structures, the isolation of a new contribution, careful verification, proper citation, and time for reflection. Companies, however, optimize for speed and public benchmarks: the faster a model "closes" a known problem, the higher its valuation in the eyes of investors and users. Such acceleration, in the laureates' view, destroys the "fertile soil" for new ideas and undermines the process of nurturing students and ideas.
Methodologically, the statement relies on observations of real incidents. Mathematicians note that AI systems often produce solutions without a full written exposition, without isolating what is genuinely new, and without taking prior work into account. This is directly linked to recent disputes over priority in problems where OpenAI allegedly used results obtained with the help of models from other laboratories without proper attribution. Unlike the traditional mathematical process, where a proof passes through peer review and discussion in the community, the machine approach leaves little room for such reflection.
Compared with previous approaches to integrating computational tools into mathematics, the current situation differs in scale and commercial pressure. Previously, computers were used for verification or for searching for counterexamples, but not for the mass production of "solutions" to known problems in record time. Parallel efforts by laboratories such as DeepMind or Anthropic also focus on benchmarks, but it is OpenAI that in recent weeks has most aggressively promoted the narrative of a "breakthrough" in the Millennium Prize Problems. This creates a competitive dynamic in which speed becomes the main advantage rather than depth of understanding.
The consequences for the field are obvious: if the community begins to orient itself toward machine solutions as the primary means of advancement, the space for slow but fundamental work will shrink. Students and young researchers may lose motivation for deep study, since the "answers" already exist in the public domain. Moreover, there arises a risk of the erosion of authorship and responsibility — who bears responsibility for an error in an AI proof if it is assembled from thousands of fragments?
The statement does not deny the usefulness of AI for mathematics, but emphasizes the need for an urgent discussion of the rules of interaction. The laureates call on the mathematical community, developers, and society as a whole to develop mechanisms that will preserve the value of conceptual understanding and proper citation. Without such mechanisms, the divergence of goals will only intensify, affecting not only mathematics but also other areas of intellectual labor.
The key question that remains open: will the community of mathematicians succeed in imposing its standards of verification and authorship on AI developers, or will commercial logic definitively redefine what counts as "progress" in fundamental science.
